GrimACE: automated, multimodal cage-side assessment of pain and well-being in mice

lab animal Article https://doi.org/10.1038/s41684-026-01695-9 GrimACE: automated, multimodal cage-side assessment of pain and well-being in mice Check for updates Oliver Sturman    1,2,3 , Marcel Schmutz    1,2,3, Tom Lorimer1,2,3, Runzhong Zhang1,2, Mattia Privitera1,2, Fabienne K. Roessler1,2, Justine Leonardi    1,2, Rebecca Waag1,2, Alina-Mariuca Marinescu    1,2, Clara Bekemeier4,5, Katharina Hohlbaum    6 & Johannes Bohacek    1,2,3 Pain and welfare monitoring is essential for ethical animal testing, but current cage-side assessments are qualitative and subjective. Here we present the GrimACE, a fully standardized and automated cage-side monitoring tool for mice, the most widely used animals in research. The GrimACE uses computer vision to provide automated mouse grimace scale (MGS) assessment together with pose estimation in a safe, dark environment. We validated the system by analyzing pain after brain surgeries (craniotomies) with head implants under two analgesia regimes. Human-expert and automated MGS scores showed very high correlation (Pearson’s r = 0.87). Both expert and automated scores revealed that a moderate increase in pain can be detected for up to 48 h after surgeries, but that both a single dose of meloxicam (5 mg/kg subcutaneuously) or three doses of buprenorphine (0.1 mg/kg) + meloxicam (5 mg/kg subcutaneuously) provide adequate and comparable pain management. Simultaneous pose estimation demonstrated that mice receiving buprenorphine + meloxicam showed increased movement 4 h after surgery, indicative of hyperactivity, a well-known side effect of opioid treatment. Significant weight loss was also detected in the buprenorphine + meloxicam treatment group compared with the meloxicam-only group. In addition, detailed BehaviorFlow analysis and automated MGS scoring of control animals suggests that habituation to GrimACE is unnecessary, and that measurements can be repeated multiple times, ensuring standardized postoperative recovery monitoring. The evaluation of pain and well-being in laboratory animals is an essential part of all ethical experimentation1,2. Although mice are the most commonly used animal model in scientific studies owing to their genetic similarities to humans and their utility in understanding various diseases and treatments3,4, accurate assessment of pain and well-being in mice is often challenging. Inadequate pain management not only raises serious ethical concerns regarding the humane treatment of animals, but also jeopardizes the validity and reproducibility of research findings5,6. With good pain and welfare monitoring protocols in place, it is possible to design appropriate analgesia regimes, detect problems before they become too severe, and define and work with humane endpoints. Moderate-to-severe pain in laboratory mice is most often related to surgical interventions, which are a cornerstone of in vivo animal research. The gold standard to assess postsurgical recovery is based on cage-side assessment7, where key behavioral parameters (for example, posture, coat condition, movement patterns and wound licking) are assessed by a trained experimenter via visual inspection2,8–10. These visual observations, which provide qualitative scores or counts for individual parameters, are well suited for rapid assessment of postsurgical recovery by personnel trained in animal experimentation or by trained animal caretakers. However, cage-side assessment is prone to bias, subjectivity and poor sensitivity to subtle alterations in well-being11. It is also widely believed 1 Laboratory of Molecular and Behavioral Neuroscience, Institute for Neuroscience, Department of Health Sciences and Technology, ETH, Zurich, Switzerland. 2Neuroscience Center Zurich, ETH Zurich and University of Zurich, Zurich, Switzerland. 3ETH Zurich 3R Hub, ETH, Zurich, Switzerland. 4 Institute of Animal Welfare Animal Behavior and Laboratory Animal Science School of Veterinary Medicine Freie Universitat Berlin, Berlin, Germany. 5 Science of Intelligence, Research Cluster of Excellence, Berlin, Germany. 6German Centre for the Protection of Laboratory Animals (Bf3R) German Federal Institute for Risk Assessment (BfR), Berlin, Germany. e-mail: ; Lab Animal Article https://doi.org/10.1038/s41684-026-01695-9 that prey animals may hide signs of pain, making it difficult for observers in close proximity to accurately assess subtle changes in well-being12. By contrast, measures such as telemetry for movement and heart rate, nest-building behavior or burrowing behavior have been shown to be more sensitive indicators of postsurgical pain and recovery, as they can reveal pain-related changes when standard cage-side assessment fails to reveal impairments13–17. However, these tests initially require surgeries to implant transmitters or habituation and then prolonged observation periods in single-housing conditions to analyze complex behaviors, rendering these tests impractical for routine use in laboratories that do not specialize in pain assessment. Over the past decade, the assessment of facial features to detect the affective component of pain has been popularized through the development of the mouse grimace scale (MGS)18,19. This approach requires minimal habituation and only brief periods of surveillance using photo or video recordings. Subsequent manual scoring assesses whether signs of pain can be detected across five facial features (orbital tightening, nose bulge, cheek bulge, ear position and whisker change), and each feature is assigned a value from 0 to 2 (0 = absent, 1 = moderate, 2 = severe). This scoring process is very labor intensive, requires highly trained experimenters18,20 and remains subject to bias21. Several groups have developed pipelines to automate (parts of) this process22–26; however, since experimental setups vary between labs, automated pipelines do not transfer well between labs and setups. Moreover, in many scenarios, the assessment of grimace scores with automated software is particularly challenging, such as when head implants or other interventions (fresh wound sites on the head with ointment from sterilization and local anesthetics) alter the images. This is particularly problematic, as craniotomies are the most commonly used surgical procedures in neuroscience research, and pain associated with craniotomies is notoriously difficult to detect using cage-side assessment27,28. MGS scores are highly sensitive to pain after craniotomies, showing that pain typically peaks 4–6 h after surgery, before resolving gradually over the course of 24–48 h (refs. 6,9,27,29). These studies also suggest that nonsteroidal anti-inflammatory drugs (NSAIDs) such as meloxicam or carprofen provide adequate analgesia27,29. Finally, going beyond classical pain assessment tools, deep behavioral profiling has recently emerged for pain detection, leveraging machine learning to extract subtle behavioral motifs from video recordings of freely moving animals30–3 (...truncated)